TM★

Print layout lab

Four ways to make a web page read like a magazine spread — same story each time, different plumbing. Auditioning for the project pages.

№ 1

Corner float, tilted wrap

The oldest trick (float) plus the newest polish (shape-outside): the text doesn't wrap the image's rectangle, it wraps its tilt.

the ad machine
fig. 1 — agentic, unsupervised

The farm had a marketing problem that looked like a farming problem. Vegetables were selling through a wholesaler at wholesale margins, the CSA had empty slots, and the person best qualified to fix it was also the person driving the tractor.

So the tractor driver built an ad machine. An agentic system that writes the campaigns, generates the creative, watches the numbers, and adjusts spend — while its operator is out planting brassicas.

The first month it ran, return on ad spend landed between four and six times. Not a demo, not a benchmark: real dollars from real neighbors buying real vegetables.

The interesting part isn't that AI wrote ads. It's the shape of the system around it — the guardrails that keep a language model from promising kohlrabi that was never seeded, the feedback loop from the store's actual inventory, the weekly cadence that matches how a farm actually operates.

№ 2

Center runaround

The true print move — image in the middle, text on both sides. CSS can't split a line around an obstacle, so this is two columns with invisible floats carving the hole; you read down the left, then down the right, exactly like a magazine.

The farm had a marketing problem that looked like a farming problem. Vegetables were selling through a wholesaler at wholesale margins, the CSA had empty slots, and the person best qualified to fix it was also the person driving the tractor.

So the tractor driver built an ad machine. An agentic system that writes the campaigns, generates the creative, watches the numbers, and adjusts spend — while its operator is out planting brassicas.

The first month it ran, return on ad spend landed between four and six times. Not a demo, not a benchmark: real dollars from real neighbors buying real vegetables.

The interesting part isn't that AI wrote ads. It's the shape of the system around it — the guardrails that keep a language model from promising kohlrabi that was never seeded, the feedback loop from the store's actual inventory, the weekly cadence that matches how a farm actually operates.

Every part of the pipeline had to earn its keep. A model that hallucinates a discount code costs actual money; a campaign that wins costs nothing to keep winning. That asymmetry is the whole design brief.

None of this is a case study from a lab. It's a half-acre business that pays its bills, run by software its owner wrote, which is either the future of small agriculture or just what happens when an engineer isn't allowed to sit still.

4–6× ROAS
fig. 2 — any rectangle, any offset
№ 3

Newsprint columns

CSS multi-column does the newspaper thing natively — and an image with column-span: all breaks the columns full-width, mid-article, like a centerfold photo.

The farm had a marketing problem that looked like a farming problem. Vegetables were selling through a wholesaler at wholesale margins, the CSA had empty slots, and the person best qualified to fix it was also the person driving the tractor.

So the tractor driver built an ad machine. An agentic system that writes the campaigns, generates the creative, watches the numbers, and adjusts spend — while its operator is out planting brassicas.

The first month it ran, return on ad spend landed between four and six times. Not a demo, not a benchmark: real dollars from real neighbors buying real vegetables.

~6k views / month
fig. 3 — spans every column

The interesting part isn't that AI wrote ads. It's the shape of the system around it — the guardrails that keep a language model from promising kohlrabi that was never seeded, the feedback loop from the store's actual inventory, the weekly cadence that matches how a farm actually operates.

Every part of the pipeline had to earn its keep. A model that hallucinates a discount code costs actual money; a campaign that wins costs nothing to keep winning. That asymmetry is the whole design brief.

The same season, an AI-built content site started ranking inside Google's AI answers — about six thousand organic views a month, pointed at the same small farm.

None of this is a case study from a lab. It's a half-acre business that pays its bills, run by software its owner wrote, which is either the future of small agriculture or just what happens when an engineer isn't allowed to sit still.

Either way, the broccoli sells out.

№ 4

Drag the picture (one flow)

A single text flow rebuilds an invisible float's shape-outside live as you drag. One flow = one wrap side: text yields toward whichever half you drop it in, and the other side of the band goes empty — CSS can't split a line around an obstacle.

drag me
fig. 4 — art direction, live

The farm had a marketing problem that looked like a farming problem. Vegetables were selling through a wholesaler at wholesale margins, the CSA had empty slots, and the person best qualified to fix it was also the person driving the tractor.

So the tractor driver built an ad machine. An agentic system that writes the campaigns, generates the creative, watches the numbers, and adjusts spend — while its operator is out planting brassicas.

The first month it ran, return on ad spend landed between four and six times. Not a demo, not a benchmark: real dollars from real neighbors buying real vegetables.

The interesting part isn't that AI wrote ads. It's the shape of the system around it — the guardrails that keep a language model from promising kohlrabi that was never seeded, the feedback loop from the store's actual inventory, the weekly cadence that matches how a farm actually operates.

Every part of the pipeline had to earn its keep. A model that hallucinates a discount code costs actual money; a campaign that wins costs nothing to keep winning. That asymmetry is the whole design brief.

The same season, an AI-built content site started ranking inside Google's AI answers — about six thousand organic views a month, pointed at the same small farm.

⚗ Back to the lab